A Data-driven Based Method To Health State Assessment And Fault Prognosis Of Key Elevator Components

As the most commonly-used public transportation in high-rise buildings,the elevator brings great convenience to people’s life and production.At the same time,the elevator accident has also caused great harm to personal safety.Normal and safe operation of the elevator is very important.With the rapid growth of elevator data,the traditional method of fault prognosis based on mathematical models has become more complex and less applicable.If the fault prognosis of elevator components can be implemented by mining and learning data rules,the accuracy,flexibility and reliability of fault prognosis will be greatly improved.The elevator component is the research object in this dissertation,and the method of health state assessment and fault prognosis to key elevator components based on elevator datasets provided by enterprise is researched.The main research work and achievements of the dissertation include:（1）The original data has multiple sources and many parameters,and there are problems of redundancy and anomalies in the data.Low-quality data will lead to lowquality prediction results.Aiming at this problem,the method of data design and data preprocessing based on multi-source data is researched in this dissertation.Data design of elevator dataset is carried out to preliminarily select available parameters through fault prognosis logic and analytical result of parameters trend.According to the characteristic of elevator data,data cleaning,data selection,data representation and data transformation are adopted,the preprocessing of elevator data is implemented.（2）The supervised learning algorithm needs to learn the data containing features and label,and then build a fault prognosis model.Aiming at the problem of label deficiency in the original data,A construction method of health indicator based on multiple linear regression and fault similarity distance is researched in ths dissertation.At the same time,in order to select key features,the analytical method of parameters correlation based on Pearson correlation coefficient is researched.The fault similarity distance is obtained by calculating the deviation of the data between the current moment and the fault moment.The multiple linear regression algorithm is used to optimize the fault similarity distance by synthesizing the remaining available features,thus constructing a quantitative indicator for evaluating the health state of elevator components.The Pearson correlation coefficient is utilized to calculate the correlation between parameters and the fault,and select key parameters related to the fault,which lays a foundation for the fault prognosis of elevator components.（3）Aiming at the problem of limited prognostic performance to the traditional fault prognosis model,the construction of fault prognosis model to elevator component based on Bidirectional Long Short-Term Memory（Bi LSTM）and attention mechanism（AM）is researched in this dissertation.Bi LSTM is utilized to process and learn long-time series data,AM is applied to optimize feature weights and select key features.The effectiveness and applicability of the fault prognosis method based on Attention Bidirectional Long Short-Term Memory（ABi LSTM）are verified through three experiments on the elevator dataset.（4）To fit the need of fault prognosis to elevator component,a fault prognosis system to elevator component is developed in this dissertation through tools and frameworks such as Python and Django web based on the method of fault prognosis to elevator component.The prognosis system includes functions such as data preprocessing,health indicator construction,analysis of parameters correlation,training and management of prognosis model,and fault prognosis,which inplement the practical application of the fault prognosis method.The method of health state assessment and fault prognosis to key elevator components and fault prognosis system are implemented in this dissertation based on the above research work.The fault prognosis method researched makes up for the deficiency of the traditional method of fault prognosis,and has a good prognostic effect for different faults.