| Elevators have been a convenient means of transportation to travel.More and more elevators are being registered and put into use.Along with the elevator fault,maintenance problems were increasingly prominent.Recently,researchers had analyzed various operating characteristics of elevators to monitor and to decrease elevator faults.In this paper,a set of elevator working condition data acquisition systems was developed to obtain the electrical signal data generated in the process of elevator operation.Positive and negative hyper-sphere models were proposed for short-term real-time diagnosis and early warning classification of elevator faults.An improved two-dimensional convolution network was proposed to diagnose the long-term aging fault of the elevator.An improved long-term memory automatic encoder network was proposed to predict the elevator faults.Finally,the elevator fault diagnosis and early warning platform were built.The main content of the thesis was as follows:1.According to the structure of the traction elevator,the factors affecting the healthy operation of the elevator were analyzed.Compared with the vibration signals of other parts,the electrical signals generated during the operation of the elevator were less affected by the external noise.Data acquisition systems to extract the elevator operation parameter was developed.The remote communication between the hardware circuit and elevator controller was implemented by a wireless module.The collected data were extracted according to the elevator control protocol.The data was also displayed visually.2.Elevator fault may occur at any time.In this paper,the hyper-sphere model was used to determine whether the data point was in the range of the safety threshold.Based on various objective reasons,not negative samples,but a large number of positive samples could be obtained in equipment anomaly detection and fault diagnosis.The method couldn’t make full use of the negative sample to make the boundary of the hyper-sphere shift to the best position.It often led to a high error rate.A short-term fault diagnosis model of positive and negative hyper-spheres was proposed.All positive and negative samples were well distributed in the projection space.The experimental results showed that the average classification accuracy of this method could reach 98.3%.It provided elevator fault monitoring and short-term real-time diagnosis a reliable and effective method.3.The elevator operation process was composed of time series segments.The time series had the information of correlation and change.Compared with single-time data,time-series was more conducive to the identification of elevator working conditions.Based on data visualization analysis,it could be found that the curve changes of the current or active power of the same elevator in the same operation process were similar.In this paper,an improved two-dimensional convolution network method was proposed to learn and to recognize the two-dimensional images of current or active power and tags composed of up and down,load,and floor changes.The experimental results showed that the recognition accuracy of this method on the current image was 98.78%,which can effectively identify the elevator medium and long-term fault.4.To realize the early prediction of elevator fault,a network,combining long-term and short-term memory automatic encoder with a multilayer perceptron,was proposed.The maximum current,generated during the elevator start-up process,was taken as the auxiliary eigenvalue to jointly input into the network for learning.The reconstruction error of long-term and short-term encoder was decreased in predicting the upcoming events.The simulation results showed that the average accuracy of fault early warning was 85.3%.5.In this paper,the elevator fault diagnosis and early warning platform were built to realize the basic information management of the elevator.Image processing,data mining,was used for the elevator fault monitoring and early warning.Real-time decision-making,and scheduling of elevator maintenance were carried out. |
