The Research And Application Of AIS-HMM Hybrid Model In Mechanical Fault Diagnosis

With the accelerated process of industrialization,machinery has been widespread and mechanical structure is more complex.As the key equipment,machinery plays a key role in enterprises.Frequent mechanical failure will bring economic losses even more serious consequences to the enterprise.Due to the special role of machinery in the enterprise,mechanical fault diagnosis technology has become a hot topic today.The traditional diagnostic techniques include noise monitoring,oil monitoring,nondestructive testing,vibration monitoring,performance trend analysis and so on.Smart technology in fault diagnosis is also widely used,such as artificial neural networks,fuzzy logic,genetic algorithms,etc.The diagnostic technology has played an important role in the diagnosis of mechanical failure.But the diagnostic capabilities of a single diagnostic technology is weak and with poor accuracy.With the complexity of mechanical failure,overlap of the fault signals,simultaneous occurrence of multiple faults and the uncertainty of the fault brought great difficulties to the fault diagnosis.To solve these problems,this paper proposes a new idea,which is about mixing Artificial Immune System(AIS)and Hidden Markov Models(HMM)in mechanical fault diagnosis system.In the hybrid model,complement of AIS and HMM is fully used.For the mechanical systems which run more smoothly,the accumulation of information is very limited and there are few failure samples,so on-line monitoring by AIS can effectively identify the working status of machinery and improve the online performance diagnostic methods.For the non-stable mechanical systems,HMM can be applied to the diagnosis system.With the strong ability of pattern classification,HMM can obtain the parameters from the same training samples at the time of identifying the fault and training.Each HMM has a corresponding pattern,when learning a new model,simply modify the HMM class without changing the other HMM can effectively improve the efficiency of the diagnostic system.With two complementary capabilities of HMM and AIS,the hybrid model can improve the accuracy of fault diagnosis and capacity of learning to adapt to complex situations,and thus discover the fault more promptly.This paper established a two-tier hybrid model,that is HMM layer and the mixed layer.The main function of layer HMM is the application of the powerful capabilities of modeling of time series.The training of HMM alignment algorithm based on the sample failure can obtain the likelihood of a fault condition.The main function of the mixed layer is to modify and correct the results derived from the HMM layer.First,AIS clonal selection algorithm is applied to evolve the antibodies,and then through the HMM Baum-Welch algorithm antibodies evolve again.In the hybrid model,the HMM is the main recognition.The recognition rate used in the affinity of the model played a guiding role in HMM training samples,combining stochastic search and the convergence of iterative accelerating,and expanded structure and affinity of the antigen antibody.