Design and development of a generic prognostics simulator

Abstract

Industries are transforming in the fourth industrial revolution, from traditional manufacturing and industrial practices to smart manufacturing with the assistance of modern smart technology. Prognostics is a key science in industry 4.0 which assists in conditionbased maintenance of industrial assets. Hence researchers, scientists and different industrial organisations are finding different ways to implement prognostics effectively. Prognostics has evolved as an essential tool for the estimation of the Remaining Useful Life of the critical components of several industrial assets. Prognostics requires a large amount of life data of the component or system for the accurate estimation of the remaining useful life of that component or system. However, the data collection process used for prognostics is a cumbersome process which consists of operating the asset from healthy condition to failure. It is not advisable to run the asset until failure because industrial assets are expensive. Also running the asset till failure is a destructive and time-consuming process. This situation creates a problem for the data collection process in prognostics and health management. This thesis focuses on resolving the scarcity of industry-grade prognostics data. For this purpose, a mechanism which can generate prognostics data without need of running the actual machine has been designed and developed. This mechanism uses the historical data of the machine for which the user wants to generate data. This historical data is condition monitoring data of that machine. To resolve the problem of data scarcity, a mechanism is required, which can provide the prognostics data similar to the actual data. For this purpose, this thesis proposes a solution named as a Generic Prognostics Simulator. GPS uses the simulation process for new data generation. Primary factors considered while developing this mechanism is; its ability to generate the degradation data of several mechanical components. This simulator uses the historical data of asset to generate new data sets using several features of that data. A generic algorithm is developed for the processing of GPS, i.e. to simulate the datasets. The basic need for massive prognostics data can be satisfied using GPS without running the actual component until failure. Also, industrial organisations can save a large amount of cost and time using GPS for the data generation process GPS enables data generation from historical condition monitoring data which resolves massive data requirement of industries. It also assists in training the latest prognostics models developed by academic researchers on the industry-grade data generated by GPS. Many research students struggle generating new models for prognostics due to lack of data; GPS can resolve this problem. The validation of the data generation process with the help of GPS using historical condition monitoring data consisting of wear and vibration data of a milling cutter has been done. The detailed results are discussed in chapter number five. Results show the similarity between actual datasets and simulated datasets which is illustrated using graphical representations of simulated data with the original datasets.