The vibration faults of steam turbine generator sets are generally not single faults. There are often several faults occurring at the same time. However, the diagnosis of multiple faults has not been solved very well. The reasons are: 1) The system is complicated, and the research on multiple faults is not deep enough. 2) In the actual project, multiple units operate at the same time, affect each other, and there are many opportunities for multiple faults. Diagnostic analysis is more difficult. 3) Diagnostic methods are not mature enough. Due to limitations of rules and knowledge bases, expert systems and other artificial intelligence methods are more difficult. The processing power is not in the heart. In the neural network diagnosis technology, if the combination of various faults is used as the training sample, the network scale will be very large, the inductive association ability of the network will be greatly reduced, or even completely failed. The probabilistic causal model proposed in this paper The genetic algorithm combines the diagnostic method, the likelihood function of the probabilistic causal model is used as the fitness function of the genetic algorithm, and can diagnose the multi-fault of the turbine generator set. 1 Probability causal model A simple diagnosis problem can be expressed as P= (D, M, C, M), where D = d1, d2,. ", dn is the finite non-empty m1, m2, ..., mn is the finite non-empty set of indications; CG DXM is the ordered relational subset defined on DXM; the solution of the MGM as the known symptom set problem is It is known that the most likely hypothesis Dj under the symptom set M, that is, the fault set with the highest prior probability between all possible hypothetical solutions. Therefore, a diagnostic maximum likelihood problem for the probabilistic causal model needs to understand the following definitions and Hypothesis 1.1 Definitions and assumptions make dfD the cause event, m/6M is the result event, then define: Diagnostic hypothesis (on nd;) A (eight di), that is, all faults in Di exist, while other faults do not exist. .
The signs of existence are known to be M+ = (A m,) A mM (Am,), ie all signs in M ​​are present, while other signs are absent.
In the case where the known di exists, the probability that di causes m, (m,:di occurs) is P(m., * for all cause events (faults) diZD, and its prior probability is known to be 0.
Steady-state reproduction without re-stringing: When forming a new generation of groups, the individuals are not repeated, that is, before an individual is added to a new generation of groups, it is checked whether the individual and the existing individuals in the group are duplicated. If it is repeated, it will obviously improve the behavior of the genetic algorithm, but it adds computation time.
Lonely - the two of the two fathers of the body of the weeping body to replace her steam turbine power generation several fault genetic algorithm new method bookmark2 in the genetic algorithm applied to the fault diagnosis of steam turbine generator sets, due to steady-state reproduction, how many generations of individuals each time The number of surrogate individuals is not easy to determine. Generally, the steady-state breeding selection method without re-stringing is adopted, which effectively avoids the occurrence of "premature" phenomenon and improves the correct rate of fault diagnosis.
The new method is an organic combination of probabilistic causal models and genetic algorithms. In the case of known signs, the probability causal model gives a function describing the probability of occurrence of a hypothesis Di, and it is used as the fitness function of the genetic algorithm. The fault diagnosis of the steam turbine generator is converted into the optimal solution problem. For the probabilistic causal model of fault solving, first determine the fault set D of the fault diagnosis problem, the prior probability Pi, the symptom set M, the causal strength C between the fault and the feature, etc. Table 1 is the prior probability of different faults, Table 2 The causal strength Cj between the fault D and the symptom M, wherein drdn is a few common faults of the steam turbine generator set; m8 is the eight characteristic frequencies of the spectrum, respectively 0. 01~0.39/(/for the steam The rotational frequency of the wheel generator set), 0.40~0.49/, Q 50/, 0.51~0.99/, 1.0/, 20/, 3.0~5.0/, >5.0/; Pi is a variety of faults determined based on long-term accumulated experience. Frequency/G/ is the spectral peak energy normalization value of each characteristic spectrum when a fault occurs. When applying genetic algorithm to solve the fault diagnosis problem of steam turbine generator set, the initial population is 50, Pc is 0.6, Pm is Q01. And adopt steady-state breeding options without re-stringing Method to prevent the occurrence of "premature" phenomenon of genetic algorithm Table 1 Prior probability under different faults Table 2: Causal strength under different fault set D and symptom set Cj Note: d|d is misaligned; unbalance; bearing Block loose; bearing and bearing bush loose; rotor rubbing; oil film whirl; thrust bearing damage; steam excitation; coupling damage; bearing torsional vibration; bearing to journal eccentricity 3 diagnosis example 1 to a power plant steamship When the generator set is analyzed, it is found that it has appeared in Q, that is, M 1, m2, m3, m4). The results of the program analysis are faults d3 and d8, that is, the bearing block loosening and the steam flow excitation occur in the steam turbine generator set. For the fault, the actual verification result is correct. If the value of Pm is between 0.3, the diagnosis result is d3 Pm has a great influence on the diagnosis result.
Example 2 When analyzing the steam turbine generator set of another plant, it is found that m5~ms are higher than the normal value, so M=(m5, m6, m7, m8) is analyzed by the program as faults d1 and d5, that is, misalignment The actual verification result of the rotor rubbing fault is correct. If Pc is between 0.3 and 0.9, the diagnosis results are d1 and d5. If Pc>0.3, the diagnosis result is erroneously visible, and Pc has a great influence on the diagnosis result.
Under the premise of selecting the population number, Pc and Pm, the diagnostic accuracy rate is >80%, indicating that the proposed diagnostic method is feasible. 4 Conclusion The new method has high reliability and practicability, but it should be noted that: a grasp the "symptom" The standard, that is, the limit of the indication value, is considered to be one of the key factors for the success or failure of the diagnosis; b. The selection of Pc and Pm has considerable ammunition. The program analysis compiled in this paper shows that different Pc and Pm pairs are diagnosed. The results have a great impact. The two parameters in this paper are selected according to experience. The specific selection criteria still need further study.
Ford Turbocharger Core,Ford Turbocharger Cartridge,Mitsubishi Chra,Td025 Core Assembly
Fengcheng Fire Turbocharger Manufacturing Co; Ltd. , https://www.fireturboleader.com