| Rotary machines are widely used in aviation, space, electric power generation, oil processing and chemical industries. Since affected by the rigorous environment and deteriorated by the aging, its operation often encounter some faults. Once these equipments are in malfunctions, it will reduce the efficiency as endanger the safety of whole production system. In order to detect the faults in time to shoot out those troubles, the methods of detecting faults should firstly be investigated. Thus, it is very important to have a deep understanding about the courses of the faults generation, when designing a high efficient rotary machine state supervise and fault detection system. Therefore, this dissertation focus on the single rotor-multistage axial compressor to conduct a series research works as follows:0 Establish the benchmarks for the compressor performance calculation model with the faults model.The standard function performance calculation is the basis for the fault analyze. After compared the merits and the shortcomings of commonly used calculation methods, this paper has utilized a full range performance calculation method which could applying to the pre-setting work points. Based on the principle of small deviation analysis, this dissertation has work out a new method for the modeling of a compressor's malfunctions. The essence of this modeling method lying on the firstly write out a matrix that represents the influence of the aerodynamics parameters, to unveil the relationship between the available data and the performance of the machine. While combine this relationship with the commonly used criteria for the detecting of machine's faults, we could grasp the characteristics of those data when a machine is in malfunctions, which will benefit for the establishing of the malfunctions detection databank that only harness thermal parameters.(D Based on an improved BP algorithm to establish the faults detection system.This paper has found that the conventional BP algorithm can not meet the needs of the faults detection system. So we use those identical faults data as the sample data when training the neural network based on the improved BP algorithm. The trained neural network can be used to detect a simple fault or a compound faults case.(D Integrated a compressor On-line performance analysis and faults detection system.By utilizing Delphi7 as the software development tool as well as applying OOP, This paper has integrated the compressor performance calculation model with a neural network faults detection model. Further more, this dissertation has probed the possibilities of applying a fuzzy neural network and the necessity of utilize an expert system that based on the knowledge of experiences. |
