Multi-scale Condition Monitoring Method Based On Big Data And Its Application

The modern industrial system has the characteristics of complex structure and high integration, which presents higher challenge on the equipment security and reliability. Therefore, condition monitoring technology has attracted more and more attention. With the development of information technology, big data in database provides an ideal platform for data-driven condition monitoring methods. In this paper, a condition monitoring method is proposed for complex multi-scale systems in which the state parameters change slowly and cumulatively on dominant scale in the long term. Based on big data, condition monitoring parameters are constructed to reflect the equipment state using multi-scale analysis. The main contributions of this dissertation can be summarized as followings:1. The characteristics of big data in modern industrial system are analyzed. With the development of measurement technology, the characteristics of big data represent more than just the volume. Sound, images and other unstructured data generated by productive processes enrich the connotation of data. As the time-frequency analysis technology matures, for instance, multi-scale analysis, which has received increasing attention. Multi-scale analysis can extract association rules, discover association between variables, and overcome the weakness of traditional data processing method which cannot balance overall trend and local similarity fluctuations.2. Scale entropy and scale exergy are proposed to reflect the system’s multi-scale characteristics. The distribution law of variables on different scales is explored. Then based on scale entropy and scale exergy, the mathematical description of research subjects is put forward to restrict the application scope of the method proposed by the paper.3. A multi-scale condition monitoring method based on big data was proposed. As the complex structures and numbers of devices are growing in modern industrial systems, requirement of condition monitoring cannot be achieved by using single traditional methods. Based on the research of different disciplines and combining algorithms, this paper presents a new condition monitoring method for the systems, in which the state parameters change slowly and cumulatively on dominant scale in the long term. The traditional model-based condition monitoring algorithm can be improved through this novel method in the aspects:(1) The systems state is divided into dynamic and steady state. The reference models are established respectively. Then, fusion residual error is obtained by fusing dynamic residual error and stead-state residual error. Thus the influence of the inaccurate dynamic state reference model can be overcome to some extent; (2) Information granulation is applied to handle with the volume data. Thus the processing of residual error can be transformed from serial into parallel, which greatly improves the computational efficiency and reduces computing time-consuming;(3) Based on the multi-scale nature of processes, multi-scale analysis is applied to fusion residual error. Noise such as unmolded factors can be eliminated and condition monitoring parameter is constructed by extracted information from different scales.4. The condition monitoring of milling roller abrasion characteristics and ash deposition detection of radiant heating surface are taken as examples to show the computational process of multi-scale condition monitoring algorithm based on big data. Abrasion index is constructed to reflect the abrasion characteristics based on the analysis of six hundred and fifty thousand data samples. Ash deposition index provide a solution to the problem of condition monitoring on deposition of water cooling wall. The algorithm presented in the paper can construct condition monitoring parameters effectively, which provides a technical basis for the control optimization and equipment predictive maintenance.